China's First Fully Domestic 100,000-GPU AI Supercluster "Dawning 8000" Debuts at WAIC 2026

On July 17, 2026, the 2026 World Artificial Intelligence Conference opened in Shanghai, where Sugon unveiled China’s first fully domestic AI supercluster with 100,000 accelerator cards—the “Sugon 8000 (Dengfeng),” which was honored as a “Treasure of the Exhibition” at the conference. The Sugon 8000 leverages “Hyper-Intelligent Fusion” technology, supporting full-precision computing from FP64 to INT8, and is tailored for applications including scientific computing, large model training, AI inference, and industrial simulation. Powered by domestically developed Hygon chips, the system interconnects 100,000 accelerator cards via scaleFabric—a native RDMA high-speed network similar to InfiniBand—and integrates the ParaStor distributed storage system to meet the high-throughput data read/write demands of large model training and scientific computing. This cluster is designed for cutting-edge fields such as quantum computing, astronomy, automotive engineering, robotics, and large AI models. For thermal management, it employs immersion-phase-change liquid cooling technology, enabling MW-level high power density deployment per rack and enhancing energy efficiency through domestically produced coolant and year-round free cooling. The Sugon 8000 features end-to-end self-developed capabilities across chips, computing, storage, networking, cooling, applications, and services. It is equipped with the Gridview ultra-large-scale cluster management platform, enabling intelligent operations and maintenance across all scenarios, and can seamlessly connect to China’s National Supercomputing Internet.